Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed personal knowledge-router that reads a small set of local wiki/memory sources and only writes back when explicitly asked.
Install this only if you want Codex to consult your local second-brain/wiki materials for broad thinking and writing questions. Review or customize the default vault paths, recall tools, and Chinese receipt template if they do not match your setup or language preference.
The skill description is broad enough to match many normal knowledge or writing requests, which can cause the router to activate outside its intended scope. In this skill, that matters because activation may trigger local file reads and optional recall tooling against personal knowledge stores, expanding data exposure and increasing the chance of unintended context access.
The 'anchor-worthy' criteria are subjective and expansive, covering broad categories like judgment, strategy, writing, and reflection without strong limiting conditions. That ambiguity can lead the agent to invoke recall tools or read local wiki files for common prompts, causing unnecessary access to personal data and behavior that the user did not clearly request.
The writeback trigger examples include only Chinese phrases alongside English, and the routing receipt template later requires Chinese labels such as '第二大脑回路'. This imposes a specific language in user-facing behavior without offering a choice or documenting an opt-in, which is a locale/language policy concern.
The required receipt format uses Chinese field labels for every 'anchor-worthy' or 'writeback-needed' answer. Because this is mandatory user-facing text and no alternative locale behavior is offered, it forces a specific language without user opt-in.
No suspicious patterns detected.
